Learn the boundaries of multilevel photon-counting optical wireless channel: A machine learning approach

Abstract

In this paper, we present a novel unsupervised machine learning (ML) algorithm to identify the constellation index of the received signal over a photon-counting optical wireless Poisson channel. The stochastic channel model is characterized by the mean value of the Poisson distributed photon number, corresponding to a distinct transmit power level. We show the condition where the Poisson channel can be well approximated with the normal distribution, and formulate the unsupervised ML problem as a clustering problem. To demodulate the transmitted symbol, a probabilistic model based on the Gaussian mixture model is utilized to learn the boundaries of constellation regions. The proposed algorithm avoids the measurement of channel parameters and is appropriate for applications with random channel conditions. Moreover, it is useful in scenarios where the direct maximization of the likelihood function is not possible. In addition, we demonstrate that the performance of the proposed technique is superior to the linear minimum mean square (LMMSE) estimation technique. Furthermore, the time complexity analysis is presented and compared with the LMMSE and K−means clustering algorithm. In addition, the generalization of the proposed algorithm is also presented.